Seizure prediction and documentation—two important problems

Epilepsy is not a disease in and of itself, but rather is a state of the brain characterised by recurrent epileptic events that occur as a result of chronic structural or functional changes in the brain. Thus, epilepsy can originate from many underlying pathological changes, including damage caused by infl ammation, trauma, and vascular events. The most frequent and obvious clinical manifestations are epileptic seizures, which are characterised by a sudden and abrupt change in the patient’s behaviour or perception, or both. Most seizures do not last longer than 1 or 2 min. Seizures can aff ect diff erent regions of the brain and develop in diff erent ways. The symptoms might be atypical or so inconspicuous that they are not even recognised as epileptic events. Many seizure types implicate brain systems that have roles in consciousness and memory processing, which partly explains why patients sometimes forget seizures or do not recognise them as events that should be documented. Because of the abnormal neuronal discharge in the neurons implicated, often only seizure-specifi c changes recorded by electroencephalography (EEG) can provide the ultimate proof of an epileptic event. Although seizures occur without warning in most patients, it is plausible that specifi c changes in brain dynamics precede the fi rst clinical and electrographic signs of a seizure. If these changes could be detected reliably, or a state of increased seizure probability could be defi ned by any other means, treatment could move away from long-term preventive use of antiepileptic drugs to EEG-triggered on-demand therapy, such as release of anticonvulsants or electrical stimulation, to prevent a seizure from occurring. Until recently, all studies about seizure prediction were based on a retrospective assessment of EEG recordings. Only a few studies have attempted to assess so-called quasi-prospective prediction algorithms—ie, those that rely only on past information— although large databases of continuous multiday recordings are available for this purpose. With the exception of a few articles, studies published so far have either neglected or failed to show that the performance of a reported prediction technique is signifi cantly better than chance, although the methods for statistical validation of prediction algorithms are well documented and straightforward to apply. In this regard, the results presented by Mark J Cook and colleagues in this issue of The Lancet Neurology are a major milestone in epileptology, showing for the fi rst time (to our knowledge) that prospective seizure prediction is possible. 15 patients were implanted with a warning device running an algorithm, which was trained on the basis of each patient’s EEG signals during a data collection phase lasting several months. In 11 patients, the required performance criteria were met, and patients advanced to an advisory phase, during which the warning device constantly indicated low, moderate, or high probability of seizure occurrence. During the fi rst 4 months of this prospective advisory phase, the warning device worked to a level better than would be expected on the basis of chance alone in eight patients. However, a note of caution is in order. Although the reported prediction algorithm worked above the level of chance (a fi nding that has not previously been shown for quasi-prospective algorithms running on continuous long-term recordings), whether this performance is also suffi cient for clinical application is unclear; this will depend on how well patients tolerate false alarms or missed seizures, and will ultimately need to be decided on an individual basis. Nevertheless, the presented results suggest that at least some patients would view the warning device as benefi cial. Cook and colleagues’ Article contains a second interesting fi nding. Seizure diaries are currently the gold standard for assessment of any type of therapeutic intervention in epilepsy, and correct seizure documentation is essential. Effi cacy measurements, anticonvulsive drug studies, and the resulting legal decisions (ie, regulatory approval of a drug) are based on the subjective counting of seizures in a diary. When comparing documentation of seizures by intracranial EEG against documentation by patients during several months, the patients’ records substantially underestimated the frequency of seizure occurrence. Furthermore, the authors reported that Published Online May 2, 2013 http://dx.doi.org/10.1016/ S1474-4422(13)70092-9